The data center outside Hangzhou does not look like a product. It looks like a bet. Rows of servers hum where a shopping festival used to be the only thing that mattered, and the people running them are no longer measured by sales but by how much compute they can hand to someone else's model.
In early 2025, Alibaba said it would spend at least 380 billion yuan (about $53 billion) on cloud computing and artificial intelligence (AI) infrastructure over the following three years. The number was not a forecast. It was a commitment, and it exceeded everything the company had spent on AI and cloud across the previous decade combined.
Why the money is moving now
For most of its life, Alibaba treated cloud as the plumbing behind its retail empire. That changed when generative AI turned compute into the product itself. Training and serving large models requires graphics processing units (GPUs), high-bandwidth networks, and data centers built for steady, massive workloads rather than spiky shopping traffic.
The company's own leadership framed the shift plainly. At its annual Apsara Conference in September 2025, Alibaba Cloud chairman and CEO Eddie Wu said demand for AI infrastructure had "far exceeded our expectations," and signaled the group would spend even more than already promised. The 380 billion yuan figure had first been announced in February 2025 alongside full-year results.
What the capital expenditure actually covers
This is capital expenditure (资本开支), not a marketing budget. The pledge points at three physical layers:
- Land and buildings. New data centers, including the company's first in Brazil, France, and the Netherlands, plus expansions in Mexico, Japan, South Korea, Malaysia, and Dubai. By late 2025 Alibaba Cloud said it operated across 91 availability zones in 29 regions.
- Silicon and servers. GPU clusters and the supporting hardware needed to train and serve large models at scale.
- Network and energy. The interconnect and power capacity that determine whether a cluster is usable, not just present.
None of this shows up in a consumer app. It shows up as capacity that other companies rent.
The competitive backdrop
Alibaba is not running this race alone. Reuters reported in January 2025 that ByteDance had earmarked more than 150 billion yuan in capital expenditure for the year, much of it aimed at AI. Tencent and Baidu have made similar moves. The pattern across Chinese internet giants is consistent: when model quality depends on raw compute, the firms with the deepest balance sheets build the factories.
Alibaba's pitch is that owning the stack — from the Qwen open-source model family to the data center it runs on — lets it serve developers cheaply and at scale. The company has said its open models have been downloaded more than 600 million times and spawned over 170,000 derivative models.
What the open-source strategy buys
Alibaba's bet assumes that giving models away for free — the Qwen family — builds a developer base that later pays for compute. It is the same logic Amazon used with AWS: make building cheap, then charge for the infrastructure underneath. The capex (资本开支) is the down payment on that loop.
This matters beyond China. Alibaba Cloud has been expanding internationally, and the 380 billion yuan plan funds overseas data centers directly. For a startup in Southeast Asia or the Middle East, the practical question is whether Alibaba's capacity is cheaper or more available than the U.S. hyperscalers' — especially where U.S. export controls limit advanced chip access.
Key takeaways
- Alibaba committed at least 380 billion yuan (about $53 billion) to AI and cloud infrastructure over three years, more than its cumulative AI-and-cloud spend in the prior decade.
- The spending is capital expenditure (资本开支) on physical capacity — data centers, GPUs, networks — not consumer product development.
- The move mirrors similar capex (资本开支) surges at ByteDance, Tencent, and Baidu as Chinese tech firms compete on compute.
What readers can do now
- Track the quarterly capex line, not the headline promise. A multi-year pledge tells you intent; the actual spend in each earnings report tells you execution.
- Watch utilization, not just capacity. New data centers only matter if customers fill them; monitor cloud revenue growth alongside the build-out.
- Compare peers. Alibaba's number is large, but ByteDance's 150 billion yuan-plus commitment shows this is an industrywide shift, not a one-company story.
Honest limitations
This article treats a public commitment as a planning signal, not a guaranteed spending path. Exchange rates move, and the RMB-to-US$figure (roughly 7.1 yuan per dollar) shifts the dollar equivalent. The pledge has no legal lock-in; Alibaba can slow or redirect spending if demand softens. Figures on rival capex come from media reports rather than filed financials and should be read as directional.
According to Li An, chief scientist at BrainNet (脑机网), "Large, multi-year infrastructure commitments like this are less about a single product and more about securing a position before demand becomes visible."